# Detecting Illegal Dumping and Encroachment Across OQGN's Gas Pipelines with AI Change Detection - Event: NestGen ’26 - Published: October 2, 2026 - Length: 30:39 - Watch: https://nestgen.org/library/watch/SdWpCgqsT8Y/ > The transcript of a recorded NestGen session, for reading or for use with an AI tool. It comes from the video's > captions; automatic captions can mishear names and terms. ## About this session Gas pipeline operators manage vast networks that are difficult to patrol consistently, leaving illegal dumping and encroachment to go undetected until it becomes a costly, sometimes dangerous problem. AI-powered change detection is giving pipeline teams a way to catch these issues automatically instead of relying on periodic manual sweeps. In this NestGen session, IBN Firnas will share how AI change detection is being used to detect illegal dumping and encroachment across OQGN's gas pipelines, and what energy teams should think through when building drone-led monitoring programs across long pipeline corridors. ## Transcript **[0:00]** Hi, welcome everyone to another exciting next gen session. You know, would love to know where are you all joining us from and what do you do? Please drop that into a chat window to make it interactive in chat. Quick fact before we begin in Oman autonomous drone inspections are being used to monitor gas pipeline right away and generate real-time alerts for issues such as illegal dumping, encroachments are also very common out there and excavation activities. Tell us in the chat whether your organization or your customers are already using drone for infrastructure monitoring like oil and gas. So today we are going to be looking at what happened when autonomous flight becomes a live information workflow that helps team act faster. Right? And for that we are going to be bringing in one of our experts today, Muhammad Khalfan who is an enterprise deployment specialist at Ibn Al Fernas Drone Center, IFDC. **[1:04]** Muhammad is an experienced UAV system engineer with a background in autonomous flight, mission planning, real-time monitoring, software and API integration and as well as a commercial drone fleet operations. In his past experience Muhammad worked at FlytBase as a lead UAV system engineer helping deploy and support autonomous drone operations. At Next Gen you know, we are excited to hear from Muhammad what Ibn Al Fernas Drone Center and OPGN have learned from running multiple autonomous inspection each day along gas pipeline right of way across Oman. You know, where they are using various AI capabilities to generate real-time alerts and from top concerns like illegal dumping, encroachment and excavation activities as we discussed. So before I hand it over to Uh, just a of housekeeping elements. Uh, for any questions that you may have, please pass that along into Q&A chat, and we would take it once the presentation is complete. This session is also getting **[2:07]** recorded and will be available for replay. Um, feel free to take a snippet of whatever you think is the best takeaway for you and post it onto social media, LinkedIn, or X. Uh, feel free to tag Muhammad and myself. With that, a very welcome very warm welcome, Muhammad, and over to you. Please share your screen and uh take us through the journey. >> Thank you, Ravi. So, uh I hope my screen is visible and let's start. Uh, just to set some context. Today, uh we're not really going to discuss about a drone or a dock. What I really want to talk about is a very simple operational problem. What happens to critical infrastructure during the hours where nobody's watching it or nobody's inspecting it. I'll take you through how we approach that problem with OQ Gas Networks. What the challenge really looked like, how AI fit into the loop, and what operationally changed, and what we learned uh while we are doing it. **[3:11]** So, thank you, Ravi, again for the quick introduction. Let me uh just give you a brief about OQ GN. OQ Gas Network operates the backbone of Oman's gas transmission system. We are talking about more than 4,000 km of pipeline infrastructure supporting customers and critical industries across the entire country. When you're dealing with a resource this fundamental, you are simply not protecting a pipeline. A serious incident at one critical point can have consequences far beyond that location, affecting almost everyone depending on it. The incident can be tiny, but the consequences may not be that small. There is another important detail. The pipeline that we are talking about here is completely underground. So, the drone flying up on the top is not really just inspecting cracks or an exposed pipeline. What we are actually interested is in the right of way, which is above the ground. We are interested in what is happening on the surface of the corridor. And **[4:11]** at this scale, the question changes. It's not just about can we actually inspect it? It becomes about how often can we inspect it to make sure that uh as soon as something relevant happens, the right people know about it. So, a right of way usually does not change very quickly. It changes slowly. Some sometimes uh there could be someone excavating near the right of way. A vehicle might enter somewhere that it should not. Waste might get dumped beside the corridor. A structure or another form of land use may start to creep in inside the right of way. Now, something to understand. None of these events necessarily need to damage the pipeline or the damage is not supposed to be done at that point. But every one of them can be an early indicator of the risk. We are not supposed to wait for the pipeline to actually get damaged for that particular event to deserve our attention. So, this could be an very ordinary event, but in the wrong **[5:12]** place around a critical infrastructure, they would actually matter. So, uh there is nothing There was nothing wrong with the fundamental process of how OQGN previously did inspection. A team was deployed on the right of way. They used to travel somewhere around 50 to 250 km just to reach that point. They used to start inspecting it and document what they actually see. Later, that information had to be compiled in a report. That report had to go from person to person or team to team, and then someone was supposed to respond to it. The problem was a patrol can only see the right of way when it is there. Now, imagine a patrol inspects the right of way at 9:00. Everything seems to be very normal. At 10:30, an excavator starts moving inside the same area. Now, the inspection at 9:00 was not a wrong. It was simply the data of that inspection now simply became old. It became irrelevant. That That creates something that I would like to call a blind window. Some the time **[6:17]** where no one was looking at that particular corridor at at that particular pipeline. The limitation that long corridors have is that they cannot be revisited quickly. They cannot be revisited that often. There was also a second delay. Even when something is observed, that still had to become a report. That information could not be used immediately to act upon it. It had to be documented and then send out send out towards the next team. The problem that we are actually trying to solve was not just about the distance. It was about the time between uh activity to happen and the right people having that information as quickly to act on that particular event. Now, uh drones was not something new to OQGN. OQGN had already been using drones as part of its operation. So, we are not introducing drones and asking can a drone inspect the right of way. We already knew that drone could provide us with that **[7:19]** visibility. The actual question was, how do we take that existing capability and make it autonomous, repeatable, and part of their daily operating routine? Instead of mobilizing a pilot to the right of way every time an aerial inspection was required, we wanted the inspection to live at that particular corridor and execute whenever an inspection was needed. Making it autonomous did not all just mean that we wanted to take people onto the drone, inspect the corridor, and people to have that process of judgment, response, and act on that particular event. Now, what you see on your screen is the core operating loop of Ocugen now life today. The missions are scheduled. The drone theory launches the aircraft, and it follows a predefined right-of-way inspection mission. But, autonomy was not simply that we **[8:21]** wanted to launch a drone without a pilot. There had to be something much more important. Uh we wanted the when the drone was flying, we wanted to analyze that AI uh video feed via AI. We wanted the alerts to be in real time. If a system identified something that potentially required attention, for example, excavation, illegal dumping, unauthorized access, encroachment, an alert was generated and sent in real time. We were not waiting for the aircraft to return, downloading hours of footage, and then to start the review process. But, AI surfacing something does not automatically make it an incident. A human remains in the loop. The operator or relevant team verifies the context, decides whether that alert is genuinely actionable, and then the operation teams can respond. Now, this had Let's talk about the few advantages this entire process gave us. First, once something was flagged and verified, the relevant frame of **[9:23]** information can move into the operational workflow, and it could become a work It had the capability of becoming a work order on the exact same day. What really enabled that speed is that everyone was looking at the same info same piece of information. The same screen, the same snapshot, the same data. The operation teams, the managers, the decision makers, the people who are responsible of issuing the work order, as well as the contractor who was responsible on resolving that particular issue. They all were working on the same visual evidence. So, instead of information moving through one team to another and potentially being interpreted differently at every stage, there was a common source of information now, around which the response can be coordinated and it makes it much more easier to move from we've detected something to we've actually acted and done something about it. Second, the scheduled missions create a continuous visual record in sense of an ongoing inspection history. Every mission adds another timestamp **[10:27]** geo-referenced record of the corridor. It creates a consistent visual evidence, a tail over the time. And that becomes particularly useful when we come back to the same location again. We see when the issue was first observed, whether it was still there on the next inspection, and eventually verify when it was actually cleared. I'll come back later to this particular in my next slide. So, then this is one of the benefits that becomes much more important that we were also able to keep people out of harm's way. Once you understand the environment that these people operate in, the these are not always easy locations to reach. We are talking about long stretches of right of way through remote desert areas. There have been multiple incidents in the past where the teams have become lost while inspecting the right of way, and a search and rescue operation was required. So, reducing the field exposure isn't simply about solving some isn't simply about saving someone a drive or just saving someone some time. Every routine **[11:33]** inspection that could be performed remotely in one less occasion is where people potentially need to travel through difficult isolated terrain unnecessarily. Of course, people still go to the site. But now they have a much better idea of where they actually need to go and why. And fourth, risk prioritized. Instead of asking somebody to give it equal attention to every minute of routine footage each and every day, the system can surface where something potentially relevant was happening. That means the human attention can go where it adds the most value. People investigate the exceptions rather than just watching watching the routine footage. And when you put all four of these together, the real objective of the entire deployment wasn't just more drone flights or more drone footage or data. It was shortening the information chain. Something happens on the right of way, it is detected, an alert is generated, the right people see the information, a human **[12:35]** verifies it, and a work order can be issued immediately. And the relevant team can go and act upon it. So, in my opinion, ultimately, the metric that mattered the most was very simple. It was how quickly can the right information reach the right people and turn it into action. So, moving on. One successful autonomous flight proves that the technology works definitely. But operations require something much more different. Monday, Wednesday, then Thursday, then Friday, and same the next week. The same corridor, the exact same mission logic again and again. That is where repeatability actually mattered. Flying a corridor once gives you an inspection. Flying that same corridor repeatedly gives you an history of what happened on that particular right of way. Let us say a system identifies an unauthorized activity or material dumped inside the corridor, the relevant OQGN team verifies it and where required a third-party contractor is instructed to remove it. Because the autonomous **[13:39]** system comes back to that same location on a later scheduled mission or a later pass, the relevant team can verify how long was that threat actually there on the right of way. How long did it remain before the contractor cleared it and we can look at the other part of the timeline also. It gives the management it it gives the team leaders opportunity to understand when was it first visible, when was it first flagged and what when did the relevant team decided that it mattered and when was the action initiated and finally when was it resolved. So, we are not only asking do we detect something, we can start asking how long was it there, how long did it take us to act and how long did it take us to close. That matters to operations because now you can verify the status of an issue on the later pass, but it also matters to the management. It creates a much clearer evidence trail around the response and the closure timeline. **[14:40]** Now, simply just instead of knowing that an issue has surfaced and eventually resolved, management can understand the entire life cycle of the threat or that issue. They can help identify where the delays occur and the support and they support better accountability in the entire process. So, once the operating model started proving itself, the next challenge that we had in front of us was scale and the deployment expanded progressively. What started around 15 km last year moved to 28 km, then 44 km, then 75 km and now we plan to complete 110 plus kilometer deployment by the end of this year. So, for me, this particular graph is not only about distance. It is about the confidence that Optical Guardian shows in the technology and shows in the team. The deployment now involves multiple docks flying eight routes a day twice daily. And not just that, **[15:44]** while they're inspecting the right of way, the same hardware, the same docks can be used for multiple use cases. It could be gas detection, it could be construction monitoring for their upcoming projects, and a lot of other things. So, the important lesson for me is that autonomous deployment should not begin with how many docks can we deploy. They should always begin with can we make one operating model work successfully, and then we can scale it. And then, scaling becomes a much more sensible conversation to take up. So, eventually, each and every technology deployment gets asked a same question. So, what? What actually improved? There are several dimensions on the slide that I would group into impact. First would be the frequency. The operating model moves the corridor away from purely periodic visibility towards scheduled autonomous inspection multiple times a **[16:48]** day. More frequent visibility meant smaller window in which relevant activity can occur without a new inspection pass. Second would be the turnaround time. A process earlier that would take up to 12 days can now be completed roughly in one. That is 12 times improvement, a 92% reduction in the time that was actually required. Third would be the worker exposure. Every routine inspection that can be performed remotely is the time that people do not necessarily need to spend driving physically covering long remote stretches of the right of way in hazardous environment. Then, there are of course secondary benefits that are reduced inspection cost, uh reduced lower emissions, traceable records, etc. The traceability point over here is particularly important. Every inspection record can carry time and and **[17:51]** information. So, exactly to what date, at what location was the record of particular evidence captured, and what happened after that. So, autonomy did not only improve one particular metric. It can change frequency, speed, safety, traceability, and economics of the entire operation. And the impact generally goes beyond flight, like lower exposure, faster detection, and frequent coverage. After working on the Ocugen deployment, the biggest outcome for us was not simply the number of flights or even one individual KPI. The more interesting change happened when the drone just stopped being a drone project and started becoming a part of Ocugen's operating infrastructure. The autonomous operation created a common information layer that supports different Ocugen functions. It could be operations, integrity checks, security, and management. The same inspection evidence can become accessible across multiple teams instead of **[18:56]** remaining one isolated function within within the drone deployment. The deployment also proved a case for autonomy at operational scale. Later, the conversation stopped uh being can autonomous drone work here and actually change to where else can we deploy this exact same operating model. And once you have built the infrastructure, the docks connectivity the docks connectivity machine planning approvals procedures data flow and operational trust, you have created a blueprint for the entire deployment that can potentially support additional corridors, use cases and much more. That same infrastructure may not only support other payloads, but also other applications like thermal inspection mapping security surveillance emergency response without you having to rebuild the entire model from zero. So, at that point, you do not really need to have a drone use case anymore. You **[19:59]** already have the infrastructure and that is why the biggest outcome was not just more flights. It was the entire new way of how the OQGN team was operating. Uh the thought that I would want to leave you guys with, we started with a very simple problem, the hours between patrol. And the time in which something could have happened on the right of way and the right people not come to know yet. Autonomy gave us a Autonomy gave us a way to make those hours smaller. Not by removing people, not by simply flying drones, but by putting repeatable inspection capability closer to the asset and getting the information to the people sooner. That has been a journey from patrol to persistent awareness and I'm open for any questions now. >> Fantastic, Muhammad. I think this is, you know, the first session in the oil and gas track that we have witnessed and started with a bang. **[20:59]** Really appreciate you taking us through the OQGN deployment. You know, it was very, very interesting to see how three dogs, eight routes, you know, just twice daily autonomous inspections can run a long pipeline right away into a continuous visual record. And the way you have implemented AI to even surface the potential concern for human verification, you know. One of the big takeaway personally for me is the the bigger shift that you mentioned, which is not more flights, but a connected operating model. >> Exactly. >> Right. And I think that support faster detection, accessible record, the kind of exposure, and the coordination across different teams. So, I think that blind window point that you explained in the early part was very insightful. I think we have few minutes and we can go over and take some questions here. I'm looking at the first one. So, it says beyond illegal dumping and encroachment is the drone inspecting the pipeline infrastructure itself? **[22:02]** Just checking. So, [snorts] eating damage and exposed pipe. >> Well, not really because as I mentioned earlier the pipeline is completely under the ground. So, because you you won't you cannot see any part or any sort of pipeline because it is buried completely under the ground. And the challenge was to keep the pipeline safe. And how do you do that was to detect the illegal excavation or any activities or any encroachments on the top of the right of way. So, of course, the entire infrastructure could So, what we've done is we've deployed the stations. We've deployed the dogs near certain stations. That gives us an opportunity to inspect that particular station as well. You have different sort of stations, but when it comes to the pipeline inspection, the main goal was not to uh uh inspect for **[23:03]** something like corrosion or cracks or something like that. >> Got it. Interesting. I think we'll move to the next one. Um we operate pipelines across multiple countries in the Middle East. Uh if we deploy this in Oman with Ebner Furnas, could we replicate it in another country with a different local partner? >> Uh yes, definitely. So, the thing you have to understand is once you have the capability to prove that this particular operating model works, replicating it becomes much more easier and much more faster. So, you have the entire journey where you know uh how exactly the deployment has to look like, how exactly you need to plan the mission, and how exactly do you need to set up your entire workflow where the human has to verify the AI alert, who is the right person for uh the team to, you know, get that alert, who needs to respond to it. And this helps what now you need to do is you just need to **[24:07]** do something of a copy-paste system. You just need to change a few variables, and you can uh this would help you to deploy the this entire same-to-same model in any of the similar zones or any of the similar pipelines. >> Fantastic. I think um you know, one of the purpose of Nshn was certainly to also cross-pollinate the business opportunities, and um uh we are glad to see that those kind of conversations are happening as part of uh the session. Moving on to the next one, uh you know, Muhammad, this is coming to what we have discussed backstage as well. Uh you know, Oman's environment is harsh where extreme heats and and unlimited connectivity, or rather limited connectivities, are the challenges, right? How does the dog hold up, and what infrastructure does it need on site to operate? >> So, definitely uh working in GCC, the environment is harsh particularly when you are working in the desert. So, you do need to come up with a few solutions to make sure that even the hardware **[25:09]** does not fail, but that I would say are operational challenges. So, particularly for us what we've done is we've identified the a window a particular window in those peak summer times where we would rather you know time the inspection in such a way that we can make sure that the entire hardware is safe. And also when we talk about limited connectivity, we've come up with a solution of something called Fernas Hub. What this exact solution does is it provides you with power, connectivity entirely and it can be moved from one point to another. And that is how we've dealt with that particular challenge for even Fernas from center. >> Yeah, I think GCC is certainly a slightly different area or region in order to address and DJI Doc 4 is trying to see how we can improve **[26:11]** that one. But yeah, moving on to the next one is you know another interesting question coming your way which is managing the gas network with the encroachment related thing. You can have a question we need to on your screen as well Muhammad, but I think I would add a little bit to it because it is not just related to how long a doc would be able to cover, but even the timeline, right? Can it operate for 24/7 if you wanted to throw some light of how you are doing that for OKU and not to have that blind window as you explained. >> So, 100% when I say that we want to avoid or what we want to eliminate the blind window, okay, it does not necessarily mean that you need to have that visibility 24/7. You just need to make sure that the inspections are frequent enough that anytime there is a activity which is actually relevant to you, you get notified about it. You get alerted about it. So, what we can do what we've actually done is we've **[27:13]** identify those particular patches of pipeline where let's say one particular part of the pipeline the activities might not be that much. So, we've done the inspection maybe once every day. But, there could be one particular part of pipeline where there is a lot of activities going on around the clock. So, we've done the inspection four times over there. So, that is how we've managed that. And as the question says that it happens in remote areas. I would like to just answer that that currently where there is one road much larger one. >> Mohammed, looks like we lost you for last 15-20 seconds. If you don't mind just covering up the last bit for your response. >> Definitely. So, I just wanted to answer the question on the screen how long a **[28:15]** stretch of pipeline can one dock cover. Currently, we are covering up to 14 km of pipeline in one go. And we also have plans to deploy multiple docks with a single drone using the drone hopping feature to make sure that in future we are also able to, you know, increase that capability much more. >> Got it. And I think we got just the last one. And looks like you have already answered part of this last question. Maybe the second part could be interesting. If you wanted to quickly respond on to that one, how quickly can it detect that something has changed? >> I would say almost immediately. The reason why I say this is let's say that you are like I said before you have an opportunity to do your missions in such a way that you can identify that how frequently the activity is happening. And now let's say you have a single part of corridor where you have flying machines four times a day. Okay, every time something **[29:20]** relevant happens the AI analyzes the video feed and it sends you an email alert immediately so that you can go you can check and you can take the right action. Maybe you can send someone to collect it at that particular point or you can also use the drone with a different type of a payload or you can attach a speaker to make sure that you can go warn that particular set and warn that particular person who is there on the site to make sure that they stay away from that particular part of pipeline. >> Fantastic. I think this was a brilliant session Muhammad. Thank you so much for taking us into this live pipeline monitoring deployment and showing us how autonomous flights become you know more timely information and actionable. So the key lesson is the value is not in just collecting the data but it's you know getting the right alert to the right team quickly enough to respond. So thank you once again Muhammad for all your warm you know experience and sharing about the OKGN case study and I really appreciate everyone for joining us. I'm sure that this session must be **[30:22]** helpful for you in terms of the autonomous inspection and measurable value in terms of the oil and gas pipeline operation. Thank you everyone for joining us. >> Thank you. Thank you everyone.